We all know the power of a great first impression. On the Dynamics 365 Virtual Agent for Customer Service team, we’ve been wondering: what’s the best greeting message to begin the conversation with a virtual agent? Because our team helps Microsoft Customer Support run one of the largest virtual agent implementations in the world (on support.microsoft.com), we were able to run some experiments with our own customers to find out.
People have a lot of opinions about what you need to tell users up front when they start conversing with a bot. In particular, many people believe you need to tell customers that they can always get to a human agent. We tested this theory and others by running customers through five quite different greetings, with over 40,000 user sessions per greeting option.
Key learning 1: Brevity is the soul of bots
What the data told us first and foremost is that brief is best. Especially when a user has already been interacting with self-help content, getting out of the user’s way fast increases engagement (interacting with the virtual agent rather than just closing it), increases problem resolution (which we measure with a survey after each session), and decreases escalation to human agents. Our best performing greeting was “I’m the Microsoft Virtual Agent. How can I help you today?”
Key learning 2: Think about penguins
Are you thinking about penguins now? When we say up front that users can “ask to talk to a person at any time” we actually prime people to escalate to a human agent. In other words, we’re putting the idea in people’s heads. Removing this suggestion decreased the escalation rate by about 7%.
When we designed the experiment, we worried that not including this phrase might decrease the number of people willing to engage with the virtual agent at all, but in fact we did not see any decrease in people’s interaction with the virtual agent. Instead, more people gave the virtual agent a chance to solve their problem instead of pulling the “talk to a person” ripcord.
So our advice is to skip the “you can talk to a person at any time” language.
Another benefit of this approach is that when users first converse with a virtual agent (before reaching a human), the virtual agent can help gather information that makes any eventual human agent contact much more efficient.
Our initial experiment on greetings was such a great learning experience, we’ll continue to do head-to-head experiments with additional greetings options, trying to find the sweet spot. We’ll also be following up with experiments about other key user experiences, and share what we learn.
You can stop thinking about penguins now.